Abstract: Using traditional label propagation algorithm in network clustering, the initial allocation of taps, the node selection and updating order of the labels are all random processes, which affect the clustering results. To overcome this problem, a new method of network clustering based on improved label propagation algorithm is proposed, in which the graph is used to describe the network and provide the basis for the network clustering. The process of the improved label propagation algorithm includes the following steps. First, the average malpractice in the network order of any two nodes with the largest common neighbor number is obtained. Second, the most similar node and neighbor node are taken as the initial core community and its initial label is assigned. Third, the random walk similarity matrix is introduced based on the node selection and the label of the node with the highest similarity. Fourth, by using the H index on the label algorithm, the update order is improved. Finally, according to the improved result, the clustering process is realized according to the clustering process of the label propagation algorithm network. Experimental results show that the proposed network clustering method has higher accuracy and stability.

Key words: computer application, node, label, propagation algorithm, network clustering, similarity